2
$\begingroup$

As far as I know, Image alignment just finds the best fit given two images of the same scene.

Whereas Image rectification, with regards to stereo vision, warps the images so that the epipolar lines are all on the same level. Essentially all common pixels between the two images are warped to be on the same y axis.

My question is, what is the difference between Image Alignment and Image Rectification, given a pair of images, when doing stereo correspondence?

If I were to find the disparity between two images, how is image alignment different to image rectification?

$\endgroup$
1
$\begingroup$

There two ways to look at this problem.

  1. In simple terms, image rectification warps both images onto a common coordinate frame by typically estimating the transformation using the epipolar geometry.

    Image alignment finds the transformation from one image to the other. It doesn't guarantee any constrains on the epipolar geometry and only one single image is warped.

  2. Image rectification might benefit from image alignment (of local patches) to obtain the correspondences. Once these are found, the epipolar lines can be forced to be parallel.

$\endgroup$
  • $\begingroup$ This makes sense to me thank you. Do you have a reference for that for me to read up on? Also, assuming a small baseline stereo image pair, could image alignment be sufficient to replace rectification? $\endgroup$ – Grim Oct 5 '15 at 0:04
-1
$\begingroup$

Well image alignment is quite general, some techniques for get the disparity use image alignment in small regions, and they can estimate the disparity by using those methods. And depending on the technique of image alignment is what you are going to obtain, if you are using the borders of the image then the process is going to find the best match for the edges no matters if the images looks very similar or not. And you have to consider that in stereo pairs you are going to have occlusions in images that can be translated to missmatching issues.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.